On September 30, 2026, DeepSeek and Huawei Technologies introduced a suite of open-source programming tools designed for Huawei Ascend AI chips. The core component is TileLang, a high-level language intended to serve as a simpler alternative to the Nvidia CUDA framework. The software package consists of six modules covering compute, communication, and programming capabilities.
DeepSeek plans to deploy at least 160,000 Huawei Ascend 950DT chips at a new data center in Inner Mongolia. The Ascend 950DT hardware is scheduled for release in Q4 2026, with partial deployment expected between late 2027 and early 2028. This move follows the optimization of the DeepSeek V4 AI model for the Ascend architecture.
The initiative aims to reduce developer switching costs and decrease reliance on Nvidia hardware. By releasing the tools as open source, the companies seek to encourage external development and testing. Nvidia CEO Jensen Huang has expressed concerns regarding the partnership, viewing the collaboration as a competitive threat to US interests.
The long-term success of this ecosystem depends on whether TileLang can match CUDA performance in real-world benchmarks. Industry observers are monitoring developer adoption of the new tools and potential shifts in market share within China as DeepSeek transitions its model training infrastructure toward domestic hardware.
